250 Startups a Year, Ten Partnerships: How Strong Cases Filter

• 250 to 10 is not a funnel
• Not better scouting, but structured openness
• Four models that work
• How the filter actually works
• What weaker programmes do instead
• The Nordic dimension
• The Structural Response
• Conclusion
• References

Screening two hundred and fifty companies to work with ten looks like ruthless selection; the interviews suggest the wide end of the funnel is doing something the narrow end cannot.

250 to 10 is not a funnel

One number from the research tends to stop people when they first read it.

“We screened around 250 startups every year and selected roughly ten for deeper pilot and co-creation programmes.”
— Innovation Community & Startup Collaboration Lead · Engineering & Consulting · Sweden

Four per cent. Read as a selection rate, it looks like an expensive way to find ten partners, and the obvious efficiency question follows immediately: why look at two hundred and fifty if you only need ten? The report’s answer is that the two ends of that process are doing different jobs. The wide end keeps the company connected to the wider ecosystem. The narrow end concentrates internal resources on a small number of relevant opportunities. Programmes that try to save money by narrowing the top of the funnel usually discover that they have removed the connection and kept the cost.

There is external evidence that selection design matters more than selection severity. Assenova and Amit analysed 8,580 startups that passed initial screening at 408 accelerator programmes across 176 countries between 2013 and 2019, and found that the benefit of participation varied substantially with programme design rather than being a general property of acceleration. A 2023 study of fifteen German corporate accelerators reached a related conclusion from the corporate side: programmes selecting for strong strategic fit with the parent company produced high growth for those startups, with trade-offs elsewhere. Corporate scouting is not an accelerator, and the settings should not be conflated, but the underlying finding transfers: what a programme does with the companies it selects, and on what basis, decides more than how many it rejects.

Bridgium’s second research report, From Discovery to Practice, draws on 48 qualitative interviews conducted between September and December 2025: 28 with innovation leaders and senior practitioners in large Nordic and European corporations, and 20 with founders and senior representatives of startups and SMEs. Its first appendix collects the practice cases: how companies actually find, filter and engage external partners when the process works.

Not better scouting, but structured openness

The intuitive fix for a poor hit rate is a better search: more events, better platforms, sharper criteria, more scouts. The report’s finding is blunt about this. Stronger collaboration does not come from more random startup exposure. It comes from structured openness and clear rules, where corporates define a problem field, invite external interpretation, filter gradually, and move only relevant cases toward pilots or implementation.

The distinction between openness and exposure is the whole argument. Exposure is being visible to many companies. Openness, as the practice cases describe it, is publishing enough about a real problem that outsiders can work out whether they are relevant, and then providing a place where that judgement can be tested before anyone commits. One respondent puts the requirement in a single line:

“A true transition happens when strategy is paired with openness — when new ideas and fresh perspectives are actively brought to the table.”
— Director, Innovation Lab · Energy Technology · Finland

Both halves of that sentence carry weight. Openness without strategy produces the noise problem the report describes at the entry stage: volume rises, relevance does not. Strategy without openness produces the opposite failure, where priorities are clear internally and invisible outside, so the only external partners who ever appear are the ones already known to someone on the team.

March’s distinction between exploration and exploitation explains why a single mechanism cannot serve both ends. Exploration rewards variety, and its value is that it surfaces options nobody specified in advance. Exploitation rewards concentration, and its value is depth on a defined problem. The 250 and the 10 are not two stages of one filter. They are two organisational logics, deliberately separated, and the ratio between them is a design choice rather than a measure of how selective the company is.

Four models that work

Appendix 1 of the report sets out four distinct models observed across the sample. They differ in what they fix and what they leave open, which turns out to be the variable that matters.

Table 1. Four practice models. Constructed from Bridgium, From Discovery to Practice (2026), Appendix 1.

Two of these deserve a second look, because they invert assumptions that most scouting programmes are built on.

The strategic domain ecosystem fixes the problem and frees the solution. In chemicals and materials, the corporation did not search for ready-made products at all. It created a focused exploration space around a shared industrial challenge, with the problem area clearly defined and possible solutions open. The respondent describes the model as articulating a shared industrial challenge while keeping potential solutions open to a broad range of contributors. This attracts specialised actors who would never have answered a product specification, because their contribution is expertise rather than a finished offering.

Knowledge sharing before pilot removes selection from the first contact entirely. In retail, the company invited startups to share what they were developing and to bring new perspectives, new ideas and new energy into the organisation, before any pilot selection was on the table.

“Around ten startups were invited for knowledge sharing, and only a smaller group later moved into internal projects where their expertise fitted existing work.”
— Director · Innovation & Product Development · Retail / Consumer Goods · Nordic Region

This creates a practical step between first contact and formal pilot selection, and it is the step most programmes do not have. Without it, the first substantive conversation is also an evaluation, which means both sides perform rather than explore. Weick’s account of sensemaking describes what the missing step provides: a setting where an ambiguous situation is interpreted jointly, before either party has committed to a reading of it.

How the filter actually works

The report is precise that filtering at the entry stage does not mean closing access too early. The stronger practice combines openness with structure, and the sequence it describes runs in six steps: open broadly, clarify the problem, invite interpretation, assess fit, filter gradually, and move only relevant cases forward.

Two things are worth noticing about that sequence. Clarification comes second, not first, which means the problem statement is refined by contact with outsiders rather than perfected before them. And interpretation is invited before fit is assessed, which reverses the usual order in which a corporate team decides internally what it wants and then tests the market against that specification.

This sequence describes the entry stage. The report also states a broader principle covering the whole path from first contact to adoption, in four steps: engage widely, explore jointly, filter gradually and embed selectively. The two are not competing lists. The six steps are the mechanics of the front door; the four steps are the discipline applied across all three stages, including what happens after a pilot succeeds.

Underneath both is a constraint that limits how many cases can be carried at once. Cohen and Levinthal showed that an organisation’s ability to use external knowledge depends on internal capacity it has already built, and that capacity is finite. Ten deep engagements is not an arbitrary number, and neither is it a target. It is roughly what an innovation function with a given headcount can carry while still being useful to each case. Widening the narrow end without adding that capacity produces more pilots and less support, which is a familiar failure mode with a good reputation.

Table 2. Recurring practices from the report against their common substitutes. Practices from Bridgium, From Discovery to Practice (2026), Appendix 1; contrasts by the Bridgium research team.

What weaker programmes do instead

Four mechanisms account for most of the difference between a programme that produces partnerships and one that produces activity.

• Events without a return path. Scouting produces contacts, and contacts arrive with no place to go. Where there is no recurring dialogue space, each conversation has to justify itself alone, and most cannot at first contact. Granovetter’s weak ties carry the useful information, but only where something exists to receive it.
• Specification instead of problem. A programme that publishes what it wants to buy hears only from companies that already sell it. A programme that publishes the problem hears from people who reframe it, which is the stated reason the strategic domain model exists.
• Selection at first contact. Where the first meeting is also the evaluation, both sides optimise for appearing suitable rather than for finding out whether they are. The knowledge-sharing step exists precisely to separate those two activities.
• No memory between rounds. The report notes that innovation memory is weak at the entry stage, with previous contacts, scouting attempts and challenge definitions not retained or searchable, so both sides restart the matching process each time. A programme without that record screens 250 companies annually and learns very little from having done it before. Burt’s account of structural holes applies to the internal case as well: the value of sitting between disconnected parts of an ecosystem depends on remembering what is on each side.

The Nordic dimension

Three of the four models in the appendix come from Nordic companies, which is worth reading as evidence rather than as coincidence of sampling.

The regional advantage is that recurring dialogue spaces are cheap to run here. Ecosystems are small, sector communities know each other, and a company that convenes a challenge programme can reach a meaningful share of the relevant actors without a large budget. Flat structures help again: the internal expert who has to give feedback in a co-design session is usually reachable, and low power distance makes that session a working conversation rather than a presentation. This is the practical reason the challenge model appears in the Finnish cases and not as an aspiration.

The constraint is arithmetic. A Nordic market does not contain 250 relevant startups a year in most sectors, which is why the Swedish case describes a broad international scouting funnel supported by external partners rather than a domestic one. Any Nordic corporate serious about the wide end of the funnel is running an international process, and that has consequences: entry points have to be legible to someone who has never heard of the company, and the ecosystem participation that generates flow has to happen outside the home market as well as inside it. The regional strength lies in the narrow end, where recurring dialogue and reachable experts do more work than budget does.

The Structural Response

Four responses follow for leaders designing or repairing a startup engagement programme.

1. Decide what you are fixing and what you are leaving open. Every working model in the appendix fixes one and frees the other: the challenge fixes the question, the domain ecosystem fixes the problem area, the knowledge-sharing format fixes only the opportunity area. A programme that fixes both the problem and the solution is running procurement, and should be resourced and named as procurement rather than as innovation.
2. Build the step between first contact and pilot selection. Knowledge exchange with no selection attached costs a few hours per company and removes the performance dynamic from the first substantive conversation. It is the cheapest structural change available and the one most consistently present in the stronger cases.
3. Size the narrow end to your capacity, not to your ambition. The number of engagements a programme can carry is set by the internal capacity to support them. Widening intake without adding that capacity converts partnerships into stalled pilots, and the resulting failure gets attributed to the startups.
4. Keep a record of the wide end. Challenge definitions, contacts, what was rejected and why, what was learned in each round. Without it, the annual screening of 250 companies is repeated from zero, and the organisation pays for exploration without accumulating any of it.

Conclusion

The ratio in the title is the least interesting thing about the case it comes from. What makes that programme work is that the wide end and the narrow end were designed to do different jobs, and neither was asked to do the other’s. Broad scouting keeps the company in contact with developments it did not anticipate. Selective concentration gives ten companies enough attention to reach a real decision. A programme with only the first produces contacts, and a programme with only the second produces a supplier list.

None of the four models in the research requires unusual budget or unusual talent. They require deciding what is fixed and what is open, creating a place where outsiders can interpret the problem before being judged on their answer, and writing down what happened so the next round starts further along. Those are organisational design choices, and they are available to any company willing to state a problem publicly.

So the question for anyone running such a programme. Of the companies you spoke to this year but did not take forward, how many could you now describe a reason for, and where is that written down?

Appendix 1 of From Discovery to Practice: How Corporate–Startup Collaboration Becomes Usable sets out the practice cases in full, with the interview evidence behind each model:
bridgium-research.eu/startup-report-2026

References

1. Bridgium Research Team (2026). From Discovery to Practice: How Corporate–Startup Collaboration Becomes Usable. Illarionova N., Verlin K., Verlin A. Report
2. Bridgium Research Team (2026). How Innovation Happens: Insights from Leading Enterprises in Times of Change. Illarionova N., Verlin K., Verlin A. Report
3. Assenova, V. A. and Amit, R. (2024). Poised for Growth: Exploring the Relationship Between Accelerator Program Design and Startup Performance. Strategic Management Journal. Wiley
4. Corporate accelerators: design and startup performance (2023). Small Business Economics. Springer
5. A meta-analysis towards the effectiveness of startup accelerators (2025). The Journal of Technology Transfer. Springer
6. March, J. G. (1991). Exploration and Exploitation in Organizational Learning. Organization Science, 2(1), 71–87. JSTOR
7. Cohen, W. M. and Levinthal, D. A. (1990). Absorptive Capacity: A New Perspective on Learning and Innovation. Administrative Science Quarterly, 35(1), 128–152. JSTOR
8. Granovetter, M. S. (1973). The Strength of Weak Ties. American Journal of Sociology, 78(6), 1360–1380. JSTOR
9. Burt, R. S. (1992). Structural Holes: The Social Structure of Competition. Cambridge, MA: Harvard University Press. Publisher
10. Weick, K. E. (1995). Sensemaking in Organizations. Thousand Oaks, CA: Sage. Publisher
11. Berger, P. L. and Luckmann, T. (1966). The Social Construction of Reality. New York: Doubleday. Publisher
12. McKinsey & Company (2020). Breaching the great wall to scale. Read